The V4 AI harness is now live, with open-weight models load balanced globally
The V4 AI harness is now live on Stack48. We also now extensively use open-weight models, load balanced globally for speed and resilience.

The V4 AI harness is now live on Stack48. If you start a free site or open the editor, this is the harness that builds and edits it.
We recently wrote about how we reworked our AI harness around what the model does best. That rework is now live for everyone who uses the AI website builder.
There is a second change alongside it. Stack48 now extensively uses open-weight models, load balanced globally for speed and resilience. Work can be served from a nearer region, and if a region or model is busy it can fail over automatically.
The reworked harness, now in your editor
The story of the rework is in the earlier V4 post. The short version is this: we rebuilt the software that sits around the model, the tools it can use, the feedback it gets, and the steps a build follows, around what today's models are already good at.
What that means in the product:
- The AI starts with your brief. You describe the site, and the harness gets on with it, whether you are starting from scratch, rebuilding an existing website, or adding pages to a complete multi-page website.
- Accurate tools and honest feedback. The platform reports what actually happened, rather than quietly patching the output.
- Checks when the AI chooses. Type-check, build, lint and form tests are there for the model to run, not forced through a long repair loop.
- One harness, three modes. Build mode just does the work. Plan mode is for bigger changes and shows a plan card with progress. Chat mode is for questions. The same harness now covers new builds and editing in the Stack48 editor.
- You can watch it work. The chat streams what the AI is reading, writing and checking in real time.
None of the platform pieces that make Stack48 a complete product have gone away. Every site still includes Stack48 Cloud for publishing and hosting, working forms that save to the site's database, and a built-in Site Admin. You can connect a custom domain, and SEO basics are generated with every page.
If you want the full account of what we reworked and why, read the V4 redesign post. This article is about the fact that it is live, and about how we now run the models behind it.
Open-weight models, load balanced globally
Stack48 now extensively uses open-weight models. In plain terms, these are models whose weights can be run in more than one place, which is what makes this kind of global load balancing possible.
We have put that to work for two reasons: speed and resilience.
- Speed. Requests can be served from a nearer region, so the AI website builder and the editor can respond more quickly than if every request had to travel to a single distant place.
- Resilience. If a region or a model is busy, the work can fail over automatically and continue elsewhere, rather than stalling the build.
We are not publishing latency figures, uptime percentages or a count of regions. The point is simpler: the live harness can send work to where it can be done well, and it can move that work if conditions change.
The harness you meet in the product does not change because a request was served from a different region. You still get the same brief-first workflow, the same tools, and the same Build, Plan and Chat modes described in the V4 rework. What changes is how that work is carried out behind the scenes.
What you'll notice
If you just want a site, here is what this means in practice.
V4 is the harness you get. A new Stack48 site is designed for its brief, not assembled from a library of pre-made sections. That is the point of an AI website builder that starts from intent. If you would rather start from a look you already like, the editor can also build a site inspired by a site you love.
The editor can keep up. Designers can still work in an AI website editor that can keep up, and larger edits can still bring in helper agents in the editor. V4 changes how that work is started, with a shared brief and one harness.
Forms and Site Admin still come as standard. Every form saves to your site's database, and is tested before the build finishes. Every site gets a built-in back office, with tables and charts, which you change by asking a small assistant.
Publishing is part of the same platform. Stack48 Cloud is included, including on the Free plan. The same is true on the Pro plan and the Agency plan for people building sites for clients. The pricing page has the current plans.
It is built for the people who actually need a site. Small businesses, tradespeople, food and hospitality businesses, health and beauty businesses and professional services can start from a short brief rather than a blank template. Web professionals and agencies can hand over a finished site rather than a locked-in mock-up. If you are comparing options, we have already set Stack48 next to other AI website builders, traditional DIY website builders and web agencies, and we have a practical look at what a small business website costs and a plain-English website guide for tradespeople. There is also a broader comparison of how AI website builders compare with traditional website builders, and a look at how Stack48 compares with a general-purpose coding agent.
Try it on your own brief
If you want to see the live harness on your own brief, start building on Stack48 for free or read how it works. You can also read about Stack48, get in touch, or browse more of this kind of writing in Stack48 Insights, including our product updates, guides and comparisons.
For the full story of the rework itself, start with We reworked our AI harness around what the model does best.
Related reading
- We reworked our AI harness around what the model does best
- Build a site inspired by any site you love
- An AI website editor for designers
- Introducing multi-agent editing in the Stack48 editor
- How AI website builders compare with traditional website builders
- Stack48 versus a general-purpose coding agent
- What a small business website costs
- A plain-English website guide for tradespeople
Frequently asked questions
- Is the V4 AI harness live for everyone?
- Yes. The reworked V4 harness is now the harness that builds and edits Stack48 sites. You can read how we reworked it in our earlier post, and start a free site to try it on your own brief.
- What are open-weight models, and how does Stack48 use them?
- Open-weight models are models whose weights can be run in more than one place, which is what makes global load balancing possible. Stack48 now uses them extensively in the live AI website builder, so work can be served from a nearer region and can fail over if a region or a model is busy.
- How does global load balancing help when I build or edit a site?
- It is there for speed and resilience. Responses can be served from a nearer region, and if a region or model is busy the work can fail over automatically. We are not publishing latency or uptime figures.
- What changed in the Stack48 AI website builder itself?
- V4 is the reworked harness we described in the earlier post. The AI gets your brief and starts building, with accurate tools, honest feedback, and checks it can run when it chooses. Build, Plan and Chat modes share the same harness. Stack48 Cloud, working forms and Site Admin still come with every site.
- Can I try the live V4 harness for free?
- Yes. You can start a free Stack48 site on the Free plan. Publishing and hosting are included through Stack48 Cloud, and you can connect a custom domain as well.

